aboutsummaryrefslogtreecommitdiff
path: root/deprecated/compare_benchmarks.py
diff options
context:
space:
mode:
Diffstat (limited to 'deprecated/compare_benchmarks.py')
-rwxr-xr-xdeprecated/compare_benchmarks.py176
1 files changed, 0 insertions, 176 deletions
diff --git a/deprecated/compare_benchmarks.py b/deprecated/compare_benchmarks.py
deleted file mode 100755
index b6a30cad..00000000
--- a/deprecated/compare_benchmarks.py
+++ /dev/null
@@ -1,176 +0,0 @@
-#!/usr/bin/python2
-#
-# Copyright 2010 Google Inc. All Rights Reserved.
-"""Script to compare ChromeOS benchmarks
-
-Inputs:
- <perflab-output directory 1 - baseline>
- <perflab-output directory 2 - results>
- --csv - comma separated results
-
-This script doesn't really know much about benchmarks. It simply looks for
-similarly names directories and a results.txt file, and compares
-the results and presents it, along with a geometric mean.
-
-"""
-
-from __future__ import print_function
-
-__author__ = 'bjanakiraman@google.com (Bhaskar Janakiraman)'
-
-import glob
-import math
-import argparse
-import re
-import sys
-
-from cros_utils import command_executer
-
-BENCHDIRS = ('%s/default/default/*/gcc-4.4.3-glibc-2.11.1-grte-k8-opt/ref/*'
- '/results.txt')
-
-# Common initializations
-cmd_executer = command_executer.GetCommandExecuter()
-
-
-def Usage(parser, message):
- print('ERROR: %s' % message)
- parser.print_help()
- sys.exit(0)
-
-
-def GetStats(in_file):
- """Return stats from file"""
- f = open(in_file, 'r')
- pairs = []
- for l in f:
- line = l.strip()
- # Look for match lines like the following:
- # METRIC isolated TotalTime_ms (down, scalar) trial_run_0: ['1524.4']
- # METRIC isolated isolated_walltime (down, scalar) trial_run_0: \
- # ['167.407445192']
- m = re.match(r"METRIC\s+isolated\s+(\S+).*\['(\d+(?:\.\d+)?)'\]", line)
- if not m:
- continue
- metric = m.group(1)
- if re.match(r'isolated_walltime', metric):
- continue
-
- value = float(m.group(2))
- pairs.append((metric, value))
-
- return dict(pairs)
-
-
-def PrintDash(n):
- tmpstr = ''
- for _ in range(n):
- tmpstr += '-'
- print(tmpstr)
-
-
-def PrintHeaderCSV(hdr):
- tmpstr = ''
- for i in range(len(hdr)):
- if tmpstr != '':
- tmpstr += ','
- tmpstr += hdr[i]
- print(tmpstr)
-
-
-def PrintHeader(hdr):
- tot_len = len(hdr)
- PrintDash(tot_len * 15)
-
- tmpstr = ''
- for i in range(len(hdr)):
- tmpstr += '%15.15s' % hdr[i]
-
- print(tmpstr)
- PrintDash(tot_len * 15)
-
-
-def Main(argv):
- """Compare Benchmarks."""
- # Common initializations
-
- parser = argparse.ArgumentParser()
- parser.add_argument('-c',
- '--csv',
- dest='csv_output',
- action='store_true',
- default=False,
- help='Output in csv form.')
- parser.add_argument('args', nargs='+', help='positional arguments: '
- '<baseline-output-dir> <results-output-dir>')
-
- options = parser.parse_args(argv[1:])
-
- # validate args
- if len(options.args) != 2:
- Usage(parser, 'Needs <baseline output dir> <results output dir>')
-
- base_dir = options.args[0]
- res_dir = options.args[1]
-
- # find res benchmarks that have results
- resbenches_glob = BENCHDIRS % res_dir
- resbenches = glob.glob(resbenches_glob)
-
- basebenches_glob = BENCHDIRS % base_dir
- basebenches = glob.glob(basebenches_glob)
-
- to_compare = []
- for resbench in resbenches:
- tmp = resbench.replace(res_dir, base_dir, 1)
- if tmp in basebenches:
- to_compare.append((resbench, tmp))
-
- for (resfile, basefile) in to_compare:
- stats = GetStats(resfile)
- basestats = GetStats(basefile)
- # Print a header
- # benchname (remove results.txt), basetime, restime, %speed-up
- hdr = []
- benchname = re.split('/', resfile)[-2:-1][0]
- benchname = benchname.replace('chromeos__', '', 1)
- hdr.append(benchname)
- hdr.append('basetime')
- hdr.append('restime')
- hdr.append('%speed up')
- if options.csv_output:
- PrintHeaderCSV(hdr)
- else:
- PrintHeader(hdr)
-
- # For geomean computations
- prod = 1.0
- count = 0
- for key in stats.keys():
- if key in basestats.keys():
- # ignore very small values.
- if stats[key] < 0.01:
- continue
- count = count + 1
- prod = prod * (stats[key] / basestats[key])
- speedup = (basestats[key] - stats[key]) / basestats[key]
- speedup = speedup * 100.0
- if options.csv_output:
- print('%s,%f,%f,%f' % (key, basestats[key], stats[key], speedup))
- else:
- print('%15.15s%15.2f%15.2f%14.2f%%' % (key, basestats[key],
- stats[key], speedup))
-
- prod = math.exp(1.0 / count * math.log(prod))
- prod = (1.0 - prod) * 100
- if options.csv_output:
- print('%s,,,%f' % ('Geomean', prod))
- else:
- print('%15.15s%15.15s%15.15s%14.2f%%' % ('Geomean', '', '', prod))
- print('')
- return 0
-
-
-if __name__ == '__main__':
- retval = Main(sys.argv)
- sys.exit(retval)